RAG (Retrieval-Augmented Generation)

Standard RAG Engineering Workflow:

1. Ingest & Chunking: Convert internal PDFs, Markdown files, or Databases into raw text. Segment these documents into overlapping chunks (e.g., 512 tokens with a 10% overlap) to preserve context.
2. Vectorization: Pass chunks through an embedding model (like OpenAI's text-embedding-3-small) to generate high-dimensional vectors.
3. Storage: Index these vectors inside a specialized Vector Database.
4. Query & Search: When a user inputs a query (e.g. "What is our company's refund policy on course cancellations?"), convert the query into a vector and perform a cosine-similarity search.
5. Prompt Construction: Inject the top-scoring text chunks directly into the LLM system instructions:

system_instruction = f"""
Use the following context to answer the user query:
---
CONTEXT:
{retrieved_chunks_text}
---
If the answer is not in the context, respond: 'I am unable to answer.'
"""

6. Execution: The LLM compiles a highly precise, hallucination-free answer based strictly on the injected context.